PRBS Stream Diagnosis

Function

  • Ascend 950PR and 950DT products: Diagnose the communication signal quality of UB direct-connect links between NPUs in a single server.
  • Other products: Diagnose the communication signal quality of hardware links by sending and receiving PRBS streams to and from the RoCE network port of the NPU.
    PRBS stream diagnosis supports the following modes:
Table 1 Diagnostic items

Item

Diagnosis Duration (Atlas A2 Training/Inference Products)

Diagnosis Duration (Atlas A3 Training/Inference Products; Ascend 950PR and 950DT Products)

Whether NPU Training or Inference Is Affected

Application Scenario

PRBS stream diagnosis

3–10 seconds

2min~10min

Yes

  • Ascend 950PR and 950DT products: Locate the signal quality issues of UB direct-connect links between NPUs in a single server.
  • Other products: Locate the signal quality issues of the RoCE network port.

Parameters

Table 2 lists only test-specific parameters. For details about other common parameters, see Common Parameters.

Table 2 Parameter description

Parameter

Description

Mandatory

[-i, --items]

Specifies the diagnosis check item.
  • PRBS stream diagnosis

Yes

[-td, --td, --trans-duration]

Specifies the duration of the PRBS stream diagnosis.

  • This parameter is available only when items is set to prbs.
  • The value ranges from 3 to 10, in seconds. If this parameter is not specified, the default value 3 is used.

No

Example

ascend-dmi -dg -i prbs -q

Example of setting PRBS stream test duration to 5:

1
2
3
4
5
6
7
8
9
[***@***]# ascend-dmi -dg -i prbs --td 5 -q
Summary:
    Arch: aarch64
    Mode: ******
    Time: 20250530-10:43:21
 
Hardware:
    prbs:
        PASS

Fault Check Items

Table 3 Fault check items

Command Output

Description

IMPORTANT_WARN

Important warning.

Bit errors are detected at the RX end and the BER is greater than 10-5. Contact Huawei technical support.

SKIP

The current product or scenario does not support this function.

PASS

The stream test is passed. The port signal quality is normal. (The BER is less than 10-5.)

FAIL

  • The stream test fails.
  • The number of bit errors reaches 67108863 for the Ascend 950PR and 950DT products and 67092480 for other products.